Post on 18-Nov-2021
TI 2008-017/1 Tinbergen Institute Discussion Paper
An Ascending Multi-Item Auction with Financially Constrained Bidders
Gerard van der Laan1
Zaifu Yang2
1 VU University Amsterdam, and Tinbergen Institute; 2 Yokohama National University, Japan.
brought to you by COREView metadata, citation and similar papers at core.ac.uk
provided by DSpace at VU
Tinbergen Institute The Tinbergen Institute is the institute for economic research of the Erasmus Universiteit Rotterdam, Universiteit van Amsterdam, and Vrije Universiteit Amsterdam. Tinbergen Institute Amsterdam Roetersstraat 31 1018 WB Amsterdam The Netherlands Tel.: +31(0)20 551 3500 Fax: +31(0)20 551 3555 Tinbergen Institute Rotterdam Burg. Oudlaan 50 3062 PA Rotterdam The Netherlands Tel.: +31(0)10 408 8900 Fax: +31(0)10 408 9031 Most TI discussion papers can be downloaded at http://www.tinbergen.nl.
An Ascending Multi-Item Auction with Financially
Constrained Bidders∗
Gerard van der Laan† and Zaifu Yang‡
February 15, 2008
∗This research was carried out while the second author was visiting the Department of Econo-
metrics, Vrije Universiteit, Amsterdam. He gratefully acknowledges financial support by the
Department and by the Netherlands Organization for Scientific Research (NWO), and by the
Ministry of Education, Science and Technology of Japan.†G. van der Laan, Department of Econometrics and Tinbergen Institute, Vrije Universiteit,
De Boelelaan 1105, 1081 HV Amsterdam, The Netherlands, e-mail: glaan@feweb.vu.nl.‡Z. Yang, Faculty of Business Administration, Yokohama National University, Yokohama 240-
8501, Japan. e-mail: yang@ynu.ac.jp
Abstract
A number of heterogeneous items are to be sold to a group of potential bidders. Every
bidder knows his own values over the items and his own budget privately. Due to budget
constraint, bidders may not be able to pay up to their values. In such a market, a Wal-
rasian equilibrium usually fails to exist and also the existing auctions might fail to allocate
the items among the bidders. In this paper we first introduce a rationed equilibrium for a
market situation with financially constrained bidders. Succeedingly we propose an ascend-
ing auction mechanism that always results in an equilibrium allocation and price system.
By starting with the reservation price of each item, the auctioneer announces the current
prices of the items in each step and the bidders respond with their demand sets at these
prices. As long as there is overdemand, the auctioneer adjusts prices upwards for overde-
manded items until a price system is reached at which either there is an underdemanded
set, or there is neither overdemand nor underdemand anymore. In the latter case the auc-
tion stops. In the former case, precisely one item will be sold, the bidder buying the item
leaves the auction and the auction continues with the remaining items and the remaining
bidders. We prove that the auction finds a rationed equilibrium in a finite number of steps.
In addition, we derive various properties of the allocation and price system obtained by
the auction.
Keywords: Ascending auction, multi-item auction, financial constraint.
JEL classification: D44.
1 Introduction
Auctions are typically the most efficient institution for the allocation of private goods and
have been long used since antiquity for the sale of a variety of items. The academic study
of auctions grew out of the pioneering work of Vickrey (1961) and has blossomed into an
enormously important area of economic research over the last 30 years. The development
in the area has been further accelerated as today governments are keen on using auctions to
sell spectrum rights, to procure goods and service, and to privatize state enterprises. Also
consumer-oriented online auctions are booming to sell virtually all sorts of commodities.
The research of the last four decades resulted in a better understanding of how the design
of auction affects its outcome and how the environments may affect the design of auctions
as well.
Standard auction theory assumes that all potential bidders have the ability to pay up
to their values on the items for sale. However in reality many buyers may be financially
constrained and may therefore not be able to afford what the items are worth to them.
Financial or budget constraints may occur in various circumstances. As stressed by Maskin
(2000) in his Marshall lecture, the consideration of financial constraints on buyers is par-
ticularly relevant and important in many developing countries, where auctions are used
to privatize state assets for the promotion of efficiency, competition and development, but
entrepreneurs may often be financially constrained. Financial constraints not only occur
in developing countries but also in developed nations. In particular, Che and Gale (1998)
have given a variety of situations where financial constraints may arise, ranging from an
agent’s moral hazard problem and business downturns, to the acquisition decisions in many
organizations which delegate to their purchasing units but impose budget constraints to
control their spending, and to the case of salary caps in many professions where budget
constraints are used to relax competition.
Financial constraints can pose a serious obstacle to the efficient allocation of the items.
For instance, financial constraints seem to have played an important role in the outcome
of auctions for selling spectrum licenses conducted in US (see McMillan (1994) and Salant
(1997)) and in European countries (see Illing and Kluh (2003)). In this paper, we study
a general model in which a number of (indivisible) items are sold to a group of financially
constrained bidders. Each bidder wants to consume at most one item. When no bidder
faces a financial constraint, the model reduces to the well-known assignment model as
studied by Koopmans and Beckmann (1957), Shapley and Shubik (1972), Crawford and
Knoer (1981), and Demange, Gale and Sotomayor (1986), among others. Each bidder has
private information about his values for the items and his budget, but these numbers are
not revealed to the other agents. In particular the auctioneer (seller) does not know the
values and budgets of the bidders and her own reservation prices is also private information
1
and is not revealed to the bidders. It is well-known (see e.g., Maskin (2000)) that even when
a single item is auctioned, it is generally impossible to have a mechanism for achieving the
full market efficiency in case bidders face budget constraints. Of course, this observation
also holds when there are multiple items for sale. Even worse, in situations that bidders
face financial constraints, a Walrasian equilibrium typically fails to exist and also allocation
mechanisms that result in an efficient allocation when bidders are unconstrained might even
fail to find a feasible allocation.
The natural question therefore is whether for situations with financial constraints a
market mechanism can be designed that at least always results in a feasible outcome
and might arrive at an outcome as efficient as possible. In this paper we first define a
rationed equilibrium for a market situation with financially constrained bidders. Succeed-
ingly we propose an ascending auction mechanism that always results in an equilibrium
allocation and price system. The auction can be seen as a modification of the well-known
ascending auction of Demange, Gale and Sotomayor (1986) for situations without financial
constraints. The auctioneer starts with the seller’s reservation price vector, that specifies
the lowest admissible price for each item, and each bidder responds with a set of items
demanded at those prices. The auctioneer adjusts prices upwards for overdemanded items
until either a price system is reached for which a set of items is underdemanded or a price
system with no underdemand and no overdemand at all. In the first case precisely one
item is assigned to a bidder that demanded that item at the previous price system, while
other bidders that demanded the item are excluded from the market for that item and thus
effectively are rationed on their demand for this item. Such a rationing can only occur
when there are several bidders with a budget equal to the selling price of the item. In case
there is neither underdemand nor overdemand anymore, an equilibrium has been reached
for all the remaining items. We prove that the auction finds a rationed equilibrium (and
thus feasible allocation) in a finite number of steps. An attractive feature of the auction is
that it only requires the bidders to report their demands at several price vectors along a
finite path rather than their values or budgets. This property is very useful and practical,
because agents generally refuse to reveal their values or budgets. This is also a reason
why English and Dutch auctions are much more popular than sealed-bid auctions. We
also derive various properties of the auction and we discuss on its efficiency. In particular,
we demonstrate that when there is no budget constraint for any bidder, the new auction
coincides with the auction of Demange, Gale and Sotomayor (1986).
We conclude this introduction by reviewing related literature. Rothkopf (1977) was
among the first to study some issues concerning sealed-bid auctions with budget con-
strained bidders. He investigated how such constraints may affect the best bids of a bid-
der. Palfrey (1980) analyzed a price discriminatory sealed-bid auction in a multiple item
2
setting under budget constraints and gave a complete characterization of Nash equilibrium
in the two items or less and two bidders or less case. Pitchik and Schotter (1988) stud-
ied the equilibrium bidding behavior in sequential auctions for the sale of two items with
budget constrained bidders. Che and Gale (1996, 1998) focused on single item auctions
with budget constraints under incomplete information. They proved that when bidders are
subject to financial constraints, the well-known revenue equivalence theorem does not hold
any more. In particular, Che and Gale (1998) provided conditions under which first-price
auctions yield higher expected revenue and social surplus than second-price auctions; see
also Krishna (2002) and Klemperer (2004). Laffont and Robert (1996) characterized an
optimal sealed-bid auction in a single item setting under financial constraints. Maskin
(2000) studied the performance of second-price auctions and all-pay auctions and pro-
posed a constrained-efficient sealed-bid auction for the sale of a single item when bidders
are financially constrained. Finally, Benoıt and Krishna (2001) investigated simultaneous
ascending auctions and sequential auctions for the sale of two items with budget con-
strained bidders. They compared the performance of both types of auctions when the two
items are complements or substitutes; see also Krishna (2002).
This paper is organized as follows. Section 2 presents the model and the notion of
rationed equilibrium. Section 3 derives a number of basic properties of over and under
demanded sets. Section 4 introduces the auction mechanism and proves its convergence.
In Section 5 it is shown that the auction results in a rationed equilibrium. Also some other
the properties of the allocation and prices generated by the auction are discussed. Section
6 concludes.
2 The Model
A seller or auctioneer has n indivisible goods for sale to a set of m financially constrained
bidders. Let N = {1, . . . , n} denote the set of the items for sale and M = {1, 2, · · · ,m}
the set of bidders. In addition to the n real items there is a dummy good, denoted by 0.
The dummy item 0 can be assigned to any number of bidders simultaneously, any real item
j ∈ N can be assigned to at most one bidder. Every bidder (he) wants to buy at most one
good. The seller (she) has for each real item j ∈ N a nonnegative integer reservation price
c(j) below which the item will not be sold. The seller will not reveal these values c(j) until
the auction starts. By convention, the reservation price of the dummy good is known to
be c(0) = 0.
Each bidder i is initially endowed with a nonnegative integer amount of mi units of
money. Further, every bidder i ∈ M attaches a (possibly negative) integer monetary
value to each item in N ∪ {0} given by the valuation function V i: N ∪ {0} → Z. Also by
3
convention, the value of the dummy item for every buyer i is known to be V i(0) = 0. The
value V i(j) of each real item j ∈ N and the amount mi are private information and thus
only known by bidder i himself. Buying an item j against price pj by bidder i yields him
a utility U i equal to
U i =
{V i(j) + mi − pj if pj ≤ mi,
−∞ if pj > mi.
That is, no bidder can afford a price above his budget mi. Since no bidder i will ever pay
more than V i(j) for any item j, his financial constraint mi is never binding when mi ≥
maxj∈N V i(j). In this case we have precisely the well known assignment market model
as studied by Koopmans and Beckmann (1957), Shapley and Shubik (1972), Crawford
and Knoer (1981), and Demange, Gale and Sotomayor (1986). In this paper we allow for
the possibility that there exist bidders i with mi < maxj∈N V i(j), i.e., there might exist
bidders whose value on some items exceeds what they can afford.
A feasible allocation π assigns every bidder i ∈ M precisely one item π(i) ∈ N ∪ {0}
such that no real item j ∈ N is assigned to more than one bidder. Note that a feasible
allocation may assign the dummy good to several bidders and that a real item j ∈ N
is unassigned at π if there is no bidder i such that π(i) = j. Let Nπ denote the set of
unassigned items at π, i.e., j ∈ Nπ if j �= π(i) for all i ∈ M . A feasible allocation π∗ is
socially efficient if∑
i∈M
V i(π∗(i)) +∑
j∈Nπ∗
c(j) ≥∑
i∈M
V i(π(i)) +∑
j∈Nπ
c(j)
for every feasible allocation π, so a socially efficient allocation maximizes the total value
that can be obtained from allocating the items over all agents. A price vector p ∈ IRn+1
yielding a price pj for every item j ∈ N ∪ {0} is feasible if pj ≥ c(j) for every j ∈ N and
p0 = 0. A pair (p, π) of a feasible price vector p and a feasible allocation π is said to be
implementable if pπ(i) ≤ mi for all i ∈ M , i.e., every bidder i can afford to buy the item
π(i) assigned to him. Given a feasible price vector p ∈ IRn+1, the demand set of bidder i is
defined by
Di(p) = {j ∈ N ∪ {0} | pj ≤ mi, V i(j)− pj = max{k∈N∪{0} | pk≤mi}
(V i(k)− pk)}.
So an item j ∈ N ∪ {0} is in the demand set Di(p) if and only if it can be afforded at p
and it maximizes the surplus V i(k)− pk over all affordable items k. Observe that for any
feasible p, the demand set Di(p) �= ∅, because p0 = 0 ≤ mi and thus the dummy item is
always affordable.
Definition 2.1 A Walrasian equilibrium (WE) is an implementable pair (p, π) such that
(a) π(i) ∈ Di(p) for all i ∈M ,
(b) pj = c(j) for any unassigned item j ∈ Nπ.
4
If (p, π) is a WE, then p is called an equilibrium price vector and π an equilibrium
allocation. From Shapley and Shubik (1972) it is well known that in a situation without
financial constraints a Walrasian equilibrium exists and that any equilibrium equilibrium
allocation is socially efficient and therefore also Pareto efficient. To find a Walrasian equi-
librium in a situation without financial constraints one can apply the auction mechanism
of Demange, Gale and Sotomayor (1986), see also Crawford and Knoer (1981), using the
notion of overdemanded set. A subset of N is overdemanded at a price vector p if the num-
ber of buyers who demand goods only from this set is greater than the number of items
in that set. Demange et al. propose an ascending auction in which the auctioneer starts
with the reservation price vector given by p0 = 0 and pj = c(j), j ∈ N . Then each bidder
is required to respond with his demand set Di(p) of most prefered items at price p. When
there is an overdemanded set of goods, the price of any item j in a minimal overdemanded
set (i.e., no strict subset of this overdemanded set is overdemanded) is increased by one and
the bidders have to resubmit their demands at this new price vector. The auction stops
as soon as there are no overdemanded sets anymore. It is well-known that the auction
stops in a finite number of steps with a minimal equilibrium price vector pmin, i.e., (i) there
exists a feasible allocation π∗ such that (pmin, π∗) constitutes a Walrasian equilibrium and
(ii) it holds that p ≥ pmin for any other equilibrium price p. Note that in the single item
case, the auction of Demange et al. (the DGS auction in short) reduces to the English
auction. These results, however, do not hold in the case of financial constraints. Namely,
the existence of an equilibrium cannot be assured anymore and also the DGS auction might
fail when some bidders face financial constraints, as illustrated in the next simple example
with only one item.
Example 1. One real item is auctioned amongst two bidders. Suppose that c(1) = 0,
V 1(1) = 5 and V 2(1) = 6. We consider three situations with respect to the budget con-
straints.
Case 1: m1 = m2 > 6. In this case the financial constraints are never binding. At the
starting price vector p = (p0, p1)⊤ = (0, 0)⊤, we have that both bidders demand the real
item 1 and N = {1} is a minimal overdemanded set. This is also the case for any p1 < 5.
When p1 = 5, then D1(p) = {0, 1} (bidder one is indifferent between the real item 1 and
the dummy item 0) and D2(p) = {1}. Thus N is not overdemanded anymore, because
at this price bidder 1 also has the dummy item in his demand set and so the number of
bidders that only have items from N is equal to one. The equilibrium price is p∗ = (0, 5)
with allocation π∗(1) = 0 and π∗(2) = 1. Observe that any p = (0, p1)⊤ with 5 ≤ p1 ≤ 6
is an equilibrium price, so the ascending auction (i.e., English auction) ends up with the
minimal equilibrium price. Also observe that the equilibrium allocation is socially efficient.
5
Case 2: m1 = 3 and m2 = 2. In this case we have that at p = (0, 2)⊤ the demands
are given by D1(p) = D2(p) = {1}, and at p = (0, 3)⊤ by D1(p) = {1} and D2(p) = {0}.
So, the latter price system yields an equilibrium in which the item is allocated to bidder 1
against price p1 = 3. Observe that the equilibrium allocation is not socially efficient. Of
course, when m1 = 2 and m2 = 3, the auction assigns the item to bidder 2 at price 3 and
the outcome is socially efficient.
Case 3: m1 = m2 = 2. We now have that at p = (0, 2)⊤ the demands are given by
D1(p) = D2(p) = {1}, and at p = (0, 3)⊤ by D1(p) = D2(p) = {0}. Now an equilibrium
does not exist and also the DGS auction fails to allocate the item to one of the bidders,
although both bidders have valuations and budgets above the zero reservation price of the
seller. At p1 = 2 the single item is overdemanded, whereas at p1 = 3 there is no demand
at all for the item. �
For the situation in Case 3 also the descending bid auction proposed in Sotomayor
(2002) fails to allocate the item to one of the bidders. Then the auction starts with high
prices and the prices of the items in a so-called underdemanded set are decreased until
there are no underdemanded sets anymore. Clearly, then at p1 = 3 the single item is still
underdemanded, but decreasing the price from 3 to 2 results in an overdemand for the
item. In Case 3, not only a Walrasian equilibrium does not exist, but also the existing
auction mechanisms fail to allocate the item. A way out of such situations that an item is
overdemanded at some price and not demanded at the next price is by allotting the item
to one of the bidders who demands the item at the highest price with overdemand, for
instance, by having a lottery between these bidders. This bidder has to pay the highest
price at which he demanded the item. Of course, allotting the item to one of these bidders
implies that the item cannot be assigned to the others who demanded also the item at the
same price. So, the auctioneer can only accept one of the bids but has to decline all other
equal bids. In the last case of Example 1, the auctioneer might accept the bid p1 = 2 of
one of the bidders, while declining the equal bid of the other. This leaves the latter bidder
with the dummy item which yields him a lower utility than the net surplus V i(1)− p1 that
he derives from item 1 against price p1 = 2. Depending on whether the item is sold to
the second or first bidder, the resulting outcome can or cannot be socially efficient. Note,
however, that even allocating the item to the first bidder gives an efficiency improvement
compared with the situation that the item remains at the seller.
It is well-known that when there are price rigidities or fixed prices, the Walrasian
equilibrium usually fails to exist. In such situations, rationing is widely used to facilitate
the allocation of goods or services; see for instance, Dreze (1975), Cox (1980), van der
Laan (1980), Kurz (1982), Azariadis and Stiglitz (1983) on divisible goods, and Talman
and Yang (2006) on indivisible goods. Typically the allocation that results from rationing
6
is not efficient. In a market situation with financially constrained bidders the necessity to
decline bids of some bidders while accepting an equal bid of one bidder also results in a
situation of rationing. After all, any bidder who leaves the auction with a net surplus lower
than the net surplus that could have been obtained from an item j that was allotted to some
other bidder feels himself a posterior rationed on the demand of such an item j. To explore
this observation, we will adapt the Walrasian equilibrium by incorporating the concept of
an allotment scheme R = (R1, · · · , Rm) where, for i ∈ M , the vector Ri ∈ {0, 1}n+1 is a
rationing vector yielding which goods bidder i can demand and for which goods offers of
bidder i will be declined. That is, Rij = 1 means that bidder i is allowed to demand good j,
while Rij = 0 means that bidder i is not allowed to demand good j ∈ N . Since the dummy
item is always available for every bidder i, we have that Ri0 = 1 for all i. Given a rationing
vector Ri with Rij = 0 for item j, the vector Ri
−j denotes the same Ri but allows bidder i
to demand item j by ignoring Rij = 0. At a feasible price vector p and an allotment scheme
R = (R1, · · · , Rm), the constrained demand set of bidder i ∈M is given by
Di(p,Ri) = {j ∈ N | Rij = 1, pj ≤ mi and
V i(j)− pj = max{k∈N∪{0} | pk≤mi and Ri
k=1}
(V i(k)− pk) }.
Now we present the solution concept of rationed equilibrium to the current model with
financially constrained bidders.
Definition 2.2 A rationed equilibrium (RE) (p, π, R) on a market with financially
constrained bidders consists of an implementable pair (p, π) and an allotment scheme R
such that
(i) π(i) ∈ Di(p,Ri) for all i ∈M ;
(ii) pj = c(j) for any unassigned item j ∈ Nπ;
(iii) If Rij = 0 for some i, then π(h) = j for some h ∈M \ {i};
(iv) If Rij = 0, then j ∈ Di(p,Ri
−j) ;
(v) If π(h) = j and there exists i �= h with Rij = 0, then mh = pj.
Conditions (i) and (ii) correspond to Conditions (a) and (b) of the definition of the Wal-
rasian equilibrium and are straightforward. In (iii)-(v) conditions on the allotment scheme
are specified. First, Condition (iii) states that rationing on an item can only occur if the
item is assigned to some bidder. So, rationing cannot occur when the item is not sold.
Condition (iv) says that any rationing is binding, i.e., if a bidder is rationed on an item,
then he will demand the item if the rationing on that item is dropped. Finally, Condition
(v) states that when some bidder is rationed on his demand of an item j, then the bidder
7
who receives the item has to pay all his money for the item and thus cannot afford a higher
price.
A rationed equilibrium (p, π, R) is said to be unrationed if no bidder is constrained on
his demand of any item, i.e., Rij = 1 for all i ∈ M and j ∈ N . An unrationed equilibrium
is just a Walrasian equilibrium.
3 Basic properties of over and underdemanded sets
To design an auction that can also deal with financially constrained bidders, we need to
find a way around the disequilibrium situation. As discussed before, a possibility to solve
the disequilibrium problem in the third case m1 = m2 = 2 of Example 1 is to sell the item
to one of the two bidders at price p1 = 2. To implement this in an ascending auction, the
auctioneer has to decide on selling the item at the previous price p1 = 2 when she observes
that the item is not demanded anymore at the price p1 = 3. The auction we design in the
next section will have this feature as one of its rules. The auctioneer starts by announcing
the seller’s reservation prices of the real items and requires the bidders to respond with
their demand sets. When there is an overdemanded set of items, the prices of the items
in a minimal overdemanded set are increased with one and the bidders are required again
to report their demand sets. This continues until there is no overdemand anymore. Then
either a situation is reached in which there is an underdemanded set of items, or there
is neither over nor underdemand. In the latter case the auction stops, in the first case
precisely one of the items in the chosen minimal overdemanded set in the previous round
is sold against its price in the previous round, after which the auction continues with the
remaining bidders by announcing again the previous round prices of the remaining items.
The notion of minimal overdemanded sets is introduced in Demange, Gale and So-
tomayor (1986) and generalized in Gul and Stachetti (2000), and the notion of underde-
manded sets can be found in Sotomayor (2002) (see also Mishra and Veeramani (2006)
and Mishra and Talman (2006)). In this section we derive several important properties of
overdemanded and underdemanded sets that will be used for proving the convergence of
the proposed auction. Let |A| stand for the cardinality of a finite set A.
For a set of real items S ⊆ N , and a price vector p ∈ IRn+1+ , define the lower inverse
demand set of S at p by
D−S (p) = {i ∈ M | Di(p) ⊆ S},
i.e., this is the set of bidders who demand only items in S. We also define the upper inverse
demand of S at p by
D+S (p) = {i ∈ M | Di(p) ∩ S �= ∅},
8
i.e., this is the set of bidders that demand at least one of the items in S. Clearly, the lower
inverse demand set is a subset of the upper inverse demand set.
Definition 3.1
1. A set of real items S ⊆ N is overdemanded at price vector p ∈ IRn+1+ if |D−
S (p)| > |S|.
2. A set of real items S ⊆ N is underdemanded at price vector p ∈ IRn+1+ if
(i) S ⊆ {j ∈ N | pj > c(j)}; and (ii) |D+S (p)| < |S|.
The definition says that a set of real items S ⊆ N is overdemanded at p if the number
of bidders who demand only items in S is strictly greater than the number of items in
S. Observe that S is a subset of real items, so any bidder i in the lower inverse demand
set does not demand the dummy item and thus has a strict positive surplus V i(j) − pj
for any item j in his demand set Di(p). An overdemanded set S is said to be minimal
if no strict subset of S is overdemanded. A set of real items S ⊆ N is underdemanded
at p if the price of every item in S is strictly greater than the seller’s reservation price
and the number of bidders who demand at least one item in S is strictly less than the
number of items in S. An underdemanded set S is said to be minimal if no strict subset
of S is an underdemanded set. We further say that an item j is overpriced if D+{j}(p) = ∅,
i.e., no bidder has item j in his demand set. Observe that S = {j} is a minimal un-
derdemanded set when item j is overpriced. So, a minimal underdemanded set S either
contains at least two (not overpriced) items, or has an overpriced item as its single element.
To give a characterization of a minimal overdemanded set, we first consider an example.
Example 2. There are three real items (n = 3) and five bidders (m = 5). Suppose
that at some p ∈ IR4+, D1(p) = {1, 2, 3}, D2(p) = {1, 2}, D3(p) = {2, 3}, D4(p) = {1}
and D5(p) = {3}. Then S = {1, 2, 3} = N is a minimal overdemanded set with D−S (p) =
{1, 2, 3, 4, 5} = M . Observe that indeed there is no strict subset of S that is overdemanded.
Further, observe that any single item in S is in the demand set of (at least) three bidders,
which number is equal to one plus the difference between the number of bidders in the
lower inverse demand set of S and the number of items in S. �
The latter observation appears to be a general property, that will play an important
role later in the proof that the auction terminates in a finite number of steps. The next
lemma states a more general property that for every nonempty subset S of a minimal
overdemanded set O at p, the number of bidders in the lower inverse demand set D−O(p)
that demand at least one item of S is at least equal to the number of items in S plus the
difference between |D−O(p)| and |O| and thus is at least one more than the number of items
in S.
9
Lemma 3.2 Let O be a minimal overdemanded set of items at a price vector p. Then,
for every nonempty subset S of O, we have
|{i ∈ D−O(p) | Di(p) ∩ S �= ∅}| ≥ |S|+ |D−
O(p)| − |O|.
Proof. Since O is overdemanded at p, the constant d = |D−O(p)| − |O| must be a positive
integer. By definition the lemma holds (with equality) for S = O. For any nonempty strict
subset S of O, define DS = {i ∈ D−O(p) | Di(p) ∩ S �= ∅}. Then we have
D−O(p) \DS = {i ∈ D−
O(p) | Di(p) ⊆ O \ S}.
Suppose to the contrary that |DS| < |S|+ d. Since 0 < |S| ≤ |O| − 1, we have that
|D−O(p) \DS| = |D−
O(p)| − |DS| > |D−O(p)| − (|S|+ d) =
= |D−O(p)| − |S| − (|D−
O(p)| − |O|) = |O| − |S| = |O \ S|.
This means that the set O \S is overdemanded, contradicting the fact that O is a minimal
overdemanded set. Hence, |DS| ≥ |S|+ d = |S|+ |D−O(p)| − |O|. �
The next corollary follows immediately.
Corollary 3.3 For every item in a minimal overdemanded set O at p, there are at least
two bidders in D−O(p) (actually the number is |D−
O(p)| − |O|+ 1 ≥ 2) demanding that item.
The properties in the next lemma and corollary are not needed for proving the conver-
gency of the auction. Nevertheless, we state these properties because they are interesting
in themselves. Considering Example 2 once more, let us assign one of the items in the
demand set D2(p) = {1, 2}, say item 1, to bidder 2. Then we claim that given this assign-
ment, there exists an assignment of all the remaining items in the set O to the remaining
bidders in D−O(p) such that (i) every bidder D−
O(p) \ {2} is assigned at most one item from
O \ {1} and (ii) every item j in O \ {1} is assigned to precisely one bidder h in D−O(p) \{2}
satisfying that j ∈ Dh(p). For example, we can assign item 2 to bidder 3 and item 3 to
bidder 5. The claim follows from the following lemma that shows a general property of
minimal overdemanded sets. The proof makes use of a well-known combinatorial theorem
of Hall (1935).
Lemma 3.4 Let O be a minimal overdemanded set at a price vector p and let T be any
subset of D−O(p) such that |T | = |O|. Then there exists an assignment of items in O to the
bidders in T such that every bidder in T is assigned at most one item, and every item in
O is assigned to precisely one bidder in T having that item in his demand set.
10
Proof. Since |O| is overdemanded, we have |O| < |D−O(p)|. Let T be any subset of
D−O(p) with |T | = |O|. We first show that for every subset S of T , the number of items
demanded by at least one bidder in S is at least as great as the number of bidders in
S. Let D(S) = ∪i∈S Di(p) be the set of items demanded by at least one bidder in S.
Suppose to the contrary that |D(S)| < |S|. Note that for every i ∈ S, Di(p) ⊆ D(S) ⊆ O
and |D(S)| < |S| ≤ |T | = |O|. This implies that D(S) is both a strict subset of the
minimal overdemanded set O and an overdemanded set, contradicting the fact that O
is a minimal overdemanded set. So we have |D(S)| ≥ |S| for all S ⊆ T and thus in
particular |D(T )| = |T | because Dh(p) ⊆ O = T for every h ∈ T ⊂ D−O(p). Now it follows
immediately from Hall’s theorem that there exists a bijective mapping π between O and
T such that ℓ ∈ Dπ(ℓ)(p) for every ℓ ∈ O. �
The lemma implies the following corollary, which shows the claim above.
Corollary 3.5 Let O be a minimal overdemanded set at a price vector p, let k be an
item in O and let h be a bidder in D−O(p) having k in his demand set. Then there exists
an assignment of items in O to the bidders in D−O(p) such that (i) item k is assigned to
bidder h, (ii) every bidder is assigned at most one item, and (iii) every item is assigned to
precisely one bidder having that item in his demand set.
Proof. Since |D−O(p)| > |O|, there exists T ⊂ D−
O(p) such that |T | = |O| and h �∈ T .
According to Lemma 3.4 there exists a bijective mapping π between O and T such that
ℓ ∈ Dπ(ℓ)(p) for every ℓ ∈ O. Let h′ be the bidder in T who is assigned item k at π, i.e.,
π(k) = h′. Now, let ρ be the assignment of items in O to the bidders in D−O(p) defined by
ρ(k) = h and ρ(j) = π(j) for any j ∈ O \ {k}, i.e., the item k that was assigned to h′ in π
is now assigned to bidder h. Clearly, the assignment ρ satisfies the conditions (i), (ii) and
(iii). �
The next result shows that the number of bidders in the upper inverse demand set
of a minimal underdemanded set is precisely one less than the number of items in the set
and that any bidder in the upper inverse demand set demands at least two items from the
minimal underdemanded set.
Lemma 3.6 Let U be a minimal underdemanded set of items at a price vector p. Then
|D+U (p)| = |U | − 1 and the demand set Di(p) of every bidder i ∈ D+
U (p) contains at least
two elements of U .
Proof. If |U | = 1, then U consists of an overpriced item and |D+U (p)| = 0. So, both
statements are true.
11
For |U | ≥ 2, denote T = D+U (p). To prove the first part, suppose |T | ≤ |U | − 2.
Then take any element k of U and denote T ′ = D+U\{k}(p). Clearly, T ′ ⊆ T and thus
|T ′| ≤ |T |. Hence
|T ′| ≤ |T | ≤ |U | − 2 = |U \ {k}| − 1
and thus U \ {k} is underdemanded, contradicting the assumption that U is a minimal
underdemanded set.
To prove the second part, suppose there is a bidder i having only one element of U
in his demand set. Let k be this element. Then T ′ does not contain bidder i ∈ T . Hence
|T ′| ≤ |T | − 1 and thus
|T ′| ≤ |T | − 1 = |U | − 2 = |U \ {k}| − 1,
showing that U \{k} is underdemanded. Again this contradicts the fact that U is a minimal
underdemanded set. �
The following lemma is given by Mishra and Talman (2006, Theorem 1) for the case
without financial constraints. In fact, the lemma is only concerned with the demand sets
of the bidders, no matter whether these demands are under financial constraints or not.
Here we provide a much simpler proof of the lemma, by using Hall’s theorem twice.
Lemma 3.7 There is a Walrasian equilibrium at p ∈ IRn+1+ if and only if at p no set
of items is overdemanded and no set of items is underdemanded.
Proof. First, let (p, π) be a Walrasian equilibrium (p, π). Clearly, at p no set of items is
overdemanded and no set of items is underdemanded.
To prove the other direction, let M1 = {i ∈ M | 0 �∈ Di(p)} and N1 = {j ∈
N | pj > c(j)}. First, consider any T ⊆ M1 and let DT = ∪i∈T Di(p). Because DT is
not overdemanded, |DT | ≥ |T |. By Hall’s Theorem, there exists a one-to-one mapping
τ : M1 → N1 such that τ (i) ∈ Di(p) for all i ∈ M1. We can extend τ to a mapping from
M to N ∪ {0} by setting τ(i) = 0 for all i �∈ M1. Next, consider any S ⊆ N1. Because S
is not underdemanded, |D−S (p)| ≥ |S|. Again by Hall’s Theorem, there exists a one-to-one
mapping ρ : N1 → M such that j ∈ Dρ(j)(p) for all j ∈ N1.
With respect to τ and ρ, denote K = {i | τ(i) ∈ N1}, L = {τ (i) | i ∈ K} and
Q = {ρ(j) | j ∈ N1 \ L} and define the mapping π: M → N ∪ {0} by
π(i) = {τ (i), for i ∈M \Q,
ρ−1(i), for i ∈ Q.
Clearly, π(i) ∈ Di(p) for all i ∈ M , and no real item is assigned by π to two different
bidders, and for every item j ∈ N1, there is a bidder i who demands the item at p and is
assigned the item. This shows that (p, π) is a Walrasian equilibrium. �
12
4 The auction design
In this section we design an ascending auction for a market with financially constrained
bidders that always results in an allocation of the items within a finite number of steps. In
the next section we discuss the properties of the resulting allocation and price system and
show that the outcome yields a rationed equilibrium. Since during the auction process the
set of bidders and the set of items are shrinking, the notions of price vector, demand set
and (minimal) overdemanded and underdemanded sets all have to be adapted accordingly.
Let C = (c(0), c(1), · · · , c(n))⊤ be the vector of the seller’s reservation prices. It should be
noted that the demand set of bidder h at stage t of the auction is given by
Dh(pt) = {j ∈ N t ∪ {0} | ptj ≤ mi, V i(j)− ptj = max{k∈Nt∪{0} | pt
k≤mi}
(V i(k)− ptk)},
where N t ⊆ N is the set of available real items and pt is the vector of prices of the items
in N t ∪ {0} at stage t.
The Ascending Auction
Step 1 (Initialisation): Set t := 1, pt := C, N t := N and M t := M . Go to Step 2.
Step 2: Every bidder i ∈ M t reports his demand set Di(pt) ⊆ N t ∪ {0}, being the
set of the most preferred items demanded by bidder i from the set N t ∪{0} at prices
pt. If there exists an underdemanded set at pt, go to Step 3. Otherwise, if there is an
overdemanded set at pt, the auctioneer chooses a minimal overdemanded set Ot of
items. Then set pt+1j := ptj+1 for every j ∈ Ot, pt+1j := ptj for every j ∈ (N t\Ot)∪{0},
M t+1 := M t and N t+1 := N t. Set t := t + 1 and return to Step 2. When there are
no overdemanded set of items and no underdemanded set of items at pt, the auction
stops.
Step 3: Let U t be a minimal underdemanded set. Then take some item k ∈ U t∩Ot−1
and bidder h ∈ {i ∈ M t | Di(pt−1) ⊆ Ot−1} such that k ∈ Dh(pt−1), but k �∈
Dh(pt) and assign item k to bidder h against price pt−1k . Set M t+1 := M t \ {h} and
N t+1 := N t \ {k}. If N t+1 = ∅, the auction stops, otherwise let pt+1j := pt−1j for all
j ∈ N t+1 ∪ {0}. Set t := t + 1 and return to Step 2.
For convenience, the parameter t in the auction will be referred to as stage t of the
auction. Note that at any stage t the price of the dummy item is equal to pt0 = c(0) = 0.
Before considering the feasibility and convergence of the auction, we first discuss the steps
in more detail and then provide an example to illustrate how the auction actually operates.
13
In Step 1, the auctioneer announces a set of items for sale and sets the starting
prices equal to the reservation prices, and a group of bidders come to bid.
In Step 2, each bidder is asked to report his demand set for the available items at
the current prices. Based on the reported demands from the bidders, the auctioneer first
checks if there is any underdemanded set of items. If so, then Step 3 will be performed.
Otherwise, the auctioneer checks whether there is any overdemanded set of items. If not,
the auction stops. According to Lemma 3.7, in this case a Walrasian equilibrium has
been reached for the remaining set of items and bidders. In case there is overdemand, the
auctioneer chooses a minimal overdemanded set of items and goes to the next stage. In
this stage the price of every item in the chosen minimal overdemanded set is increased by
one, the price of any other item remains constant and Step 2 will be performed again.
In Step 3, the auctioneer first chooses a minimal underdemanded set. Then she
selects precisely one item, say item k, that belonged to the minimal overdemanded set that
was chosen in Step 2 at the previous stage t − 1 and that also belongs to the minimal
underdemanded set at the current stage t. This item k is assigned to a bidder h satisfying
(i) who was in the lower inverse demand set of the minimal overdemanded set chosen in
stage t− 1, (ii) who demanded the item k at the previous stage t− 1, and (iii) who does
not demand item k anymore at the current stage t. This bidder h pays the price pt−1k of
item k at the previous stage and leaves the auction with the item k. The auction moves to
the next stage t+1 with the remaining items and bidders. When no real items are left, the
auction stops. Otherwise the prices of all the remaining items are set equal to the prices
in stage t− 1 and Step 2 will be performed again.
It should be noticed that in Step 3 it can never occur that there are no remaining
bidders. Clearly, this is true when the number of bidders m is larger than the number of
items n, because in Step 3 always precisely one bidder leaves with one item. When m ≤ n,
it might happen that at certain stage the auction returns from Step 3 to Step 2 with only
one bidder. Obviously then overdemand cannot occur in Step 2. In the sequel we prove that
underdemand can never occur in Step 2 when the auction returned from Step 3 in the pre-
vious stage. So, when after Step 3 the auction returns to Step 2 with precisely one bidder,
then neither underdemand nor overdemand can occur and the auction terminates in Step 2.
Example 4. Consider a market with five bidders (1, 2, 3, 4, 5) and four real items (1, 2, 3,
4). Bidders know their own values and budgets privately but the seller does not have this in-
formation. The initial endowment vector of money is given by m = (m1,m2,m3,m4,m5) =
(3, 4, 3, 5, 4) and bidders’ values are given in Table 1. The seller’s reservation price vector is
given by C = (c(0), c(1), c(2), c(3), c(4)) = (0, 2, 2, 2, 2). She will not reveal the reservation
price of any real item until the auction starts.
14
Table 1: Bidders’ values on each item.
Items 0 1 2 3 4
Bidder 1 0 4 8 5 7
Bidder 2 0 7 6 8 3
Bidder 3 0 5 5 9 7
Bidder 4 0 9 4 6 2
Bidder 5 0 6 5 4 10
This market has a unique socially efficient allocation π∗ = (π∗(1), π∗(2), π∗(3), π∗(4),
π∗(5)) = (2, 0, 3, 1, 4), which gives a total value of∑
i∈M V i(π∗(i)) = 36. Without financial
constraints, the ascending DGS auction would end at stage 7 with the minimal equilibrium
price vector p∗ = (p0, p1, p2, p3, p4) = (0, 7, 6, 8, 6) and the socially efficient allocation π∗.
The seller’s revenue generated by the auction is 27. However, in the current situation with
financial constraints, the bidders cannot afford to buy items at these minimal equilibrium
prices and so a Walrasian equilibrium does not exist. In this situation the standard as-
cending auction generates at stage 6 the price vector p = (0, 3, 3, 4, 4) (see stage 6 in Table
2 below). At this price vector there is overdemand for the items 1 and 2 (both items are
demanded by three bidders) and the prices of the items 1 and 2 are increased in the next
stage. However, at the new price vector p = (0, 4, 4, 4, 4) there is no demand anymore
for item 2, i.e., item 2 is overpriced, and the auction breaks down without reaching an
equilibrium (see stage 7 in Table 2 below). Of course, in this final stage 7 it is still possible
to allocate item 1 to the unique bidder 4 having 1 in his demand set, item 3 to the unique
bidder 2 and item 4 to the unique bidder 5. However, item 2 is not allocated and the
remaining bidders 1 and 3 don’t get any real item. The resulting allocation gives a total
value of V 2(3)+V 4(1)+V 5(4)+ c(2) = 29 and is not socially efficient. The seller’s revenue
from the auction is only 12 and her total revenue is 12 + c(2) = 14.
The new ascending auction described above yields a rationed equilibrium with both
a higher total value and higher revenue for the seller. The auction starts at stage t = 1 with
for the real items the price vector p1 = (0, 2, 2, 2, 2). Then bidders report their demand
sets: D1(p1) = {2}, D2(p1) = {3}, D3(p1) = {3}, D4(p1) = {1} and D5(p1) = {4}. The set
S = {3} is a minimal overdemanded set and the auctioneer adjusts p1 to p2 = (2, 2, 3, 2)
at stage t = 2. The price vectors, demand sets and other relevant data generated by the
auction are given in Table 2. Since pt0 = 0 for all t, these prices are deleted from the vectors
pt in the second column.
At stage 7, item 2 is overpriced and bidders 1 and 3 cannot afford any real item
because of their financial constraints. The auctioneer assigns item 2 randomly to one of
these bidders, say to bidder 1, against price p62 = 3. So bidder 1 leaves the auction with
item 2 and at the next stage 8 we have that M8 = {2, 3, 4, 5} and N8 = {1, 3, 4}. The
15
Table 2: The data generated by the auction in Example 4.
Stage pt Nt Mt D1(pt) D2(pt) D3(pt) D4(pt) D5(pt) Ot
1 (2, 2, 2, 2) {1, 2, 3, 4} {1, 2, 3, 4, 5} {2} {3} {3} {1} {4} {3}
2 (2, 2, 3, 2) {1, 2, 3, 4} {1, 2, 3, 4, 5} {2} {1, 3} {3} {1} {4} {1, 3}
3 (3, 2, 4, 2) {1, 2, 3, 4} {1, 2, 3, 4, 5} {2} {1, 2, 3} {4} {1} {4} {4}
4 (3, 2, 4, 3) {1, 2, 3, 4} {1, 2, 3, 4, 5} {2} {1, 2, 3} {4} {1} {4} {4}
5 (3, 2, 4, 4) {1, 2, 3, 4} {1, 2, 3, 4, 5} {2} {1, 2, 3} {2} {1} {4} {2}
6 (3, 3, 4, 4) {1, 2, 3, 4} {1, 2, 3, 4, 5} {2} {1, 3} {1, 2} {1} {4} {1, 2}
7 (4, 4, 4, 4) {1, 2, 3, 4} {1, 2, 3, 4, 5} {0} {3} {0} {1} {4}
8 (3, 4, 4) {1, 3, 4} {2, 3, 4, 5} {1, 3} {1} {1} {4} {1}
9 (4, 4, 4) {1, 3, 4} {2, 3, 4, 5} {3} {0} {1} {4}
auctioneer adjusts p7 to p8 = (p1, p3, p4) = (3, 4, 4), being for these three real items the
same prices as at stage 6. Then at stage 8 item 1 is overdemanded and its price is increased
to p1 = 4 at stage 9. At this stage there is neither overdemanded set nor underdemanded
set. The auction stops in Step 2 (i.e., stage 9) with, according to Lemma 3.7, a Walrasian
equilibrium for the sets of remaining items and bidders. Indeed, the auctioneer can assign
the dummy item 0 to bidder 3 (who pays nothing), and the items 1, 3 and 4 to the bidders
4, 2 and 5 respectively, against the prices p9 = (p1, p3, p4) = (4, 4, 4). The final price system
p∗ = (p0, p1, p2, p3, p4) = (0, 4, 3, 4, 4) and allocation π∗ = (π(1), π(2), π(3), π(4), π(5)) =
(2, 3, 0, 1, 4) yields a rationed equilibrium with allotment scheme R∗ = (R1, R2, R3, R4, R5),
where R∗33 = 0 and R∗i
j = 1 for all (i, j) �= (3, 3). The allocation in this rationed equilibrium
yields a total value of∑
i∈M V i(π(i)) = 35, which is slightly less than the value 36 of the
Walrasian equilibrium allocation. Even when in stage 7 item 2 should have been assigned
to bidder 3, the auction would still realise a total value of 33. In both cases the seller’s
revenue from the auction is 15, which is also equal to her total revenue, because all items
are sold. �
Now we will prove that the auction is well-designed. That is to show, all steps are
feasible and the auction stops in finitely many steps. First, observe that in each Step 3 an
item is assigned to one bidder and both the set of bidders and the set of items decrease
by one. When m ≤ n, at each stage t we have that |M t| ≤ |N t|. We will show that in
this case the auction always stops in Step 2. When m > n, then at each stage t we have
that |M t| > |N t|. In this case the auction stops either in Step 3 when N t+1 = ∅ or in
Step 2. When the auction stops in Step 3 all items are assigned sequentially in n Steps 3.
When the auction stops at some stage t in Step 2, no overdemanded or underdemanded
sets are left at the price pt, and according to Lemma 3.7, there exists a feasible allocation
πt: M t → N t such that (pt, πt) is a Walrasian equilibrium for the current sets M t and N t
at stage t.
16
Clearly, Steps 1 and 2 are feasible. So to show that the auction is well-designed,
we only need to consider Step 3. Observe that we start with all prices equal to the seller’s
reservation prices. At this starting price system there is no underdemand, because by
Definition 3.1.2 an item can only be underdemanded when its price is above its seller’s
reservation price. So, at the starting price p1 at stage t = 1, either the auction stops or
there is overdemand. In the latter case, a sequence of Steps 2 is performed with at each
stage an increase of the prices of all items in a minimal overdemanded set, until items
become underdemanded, say at stage t, and the auction goes to Step 3. So, when at some
stage t the auction goes to Step 3 for the first time, then in stage t− 1 the prices in some
minimal overdemanded set, say Ot−1, were increased. We will prove that this holds at any
stage t in which the auction goes to Step 3, i.e., when there is underdemand at some stage t,
then there was overdemand at stage t−1 and thus, when the auction reaches Step 3 at some
stage t, then at stage t− 1 the prices of the items in some minimal overdemanded set Ot−1
were increased. In Step 3 an item k in the intersection of some minimal underdemanded
set U t and the set Ot−1 is selected and assigned to a bidder h ∈ {i ∈ M t | Di(pt−1) ⊆ Ot−1}
satisfying k ∈ Dh(pt−1) \Dh(pt). The next two lemmas show that there indeed exist such
an item k and bidder h. It should be noticed that in all proofs the sets D−S (pτ ) and D+
S (pτ )
are defined with respect to the current set of bidders Mτ , for any set S ⊂ N τ and for any
τ = t− 1, t.
Lemma 4.1 Let U be a minimal underdemanded set at prices pt in some stage t and
let O be the chosen minimal overdemanded set at the previous stage t−1. Then U ∩O �= ∅.
Proof. Suppose to the contrary that U ∩ O = ∅. Since U is underdemanded at stage
t, we have that ptj > c(j) for any j ∈ U . Further, since U ∩ O = ∅, we have for any
j ∈ U that j �∈ O. Hence ptj = pt−1j and thus also pt−1j > c(j) for all j ∈ U . Since there
is no underdemand at stage t − 1, it follows that |D+U (pt−1)| ≥ |U |. Moreover, any bidder
that demands some item j ∈ U at pt−1, also demands this item at pt, because only prices
of the items in O are increased. Hence |D+U (pt)| ≥ |D+
U (pt−1)| ≥ |U | and thus U is not
underdemanded at pt, yielding a contradiction. Hence U ∩O �= ∅. �
Lemma 4.2 Let U be a minimal underdemanded set at prices pt in some stage t and
let O be the chosen minimal overdemanded set at the previous stage t−1. Then there exist
item k and bidder h satisfying the requirements of Step 3.
Proof. Since O is overdemanded at pt−1, we have |D−O(pt−1)| > |O|. Now, consider the
set S = U ∩O. By Lemma 4.1 this set is not empty. When U = O and thus S = O, then
by Lemma 3.6 there are |U | − 1 = |O| − 1 bidders demanding at least one item from U at
17
pt, because U is underdemanded at pt. So, in this case there are at least two bidders in
D−O(pt−1) not demanding any item from U = O anymore at price pt. Select h from this set
of bidders and select k from the set Dh(pt−1) (recall that this set is never empty and does
not contain any dummy item). Since Dh(pt−1) ⊆ O and for each bidder h ∈ D−O(pt−1), this
item k and this bidder h satisfy the requirements.
Next, consider the case that S is a strict subset of O. Denote H = {i ∈ D−O(pt−1) |
Di(pt−1) ∩ S �= ∅}. From Lemma 3.2 we have that
|H| ≥ |S|+ |D−O(pt−1)| − |O| ≥ |S|+ 1,
i.e., the number of bidders in D−O(pt−1) that demand an item of S at pt−1 is at least one
more than the number of items in S. Next, consider the set T = U \O. Since there is no
underdemand at pt−1 we have that
|D+T (pt−1)| ≥ |T |.
Since ptj = pt−1j for all j ∈ T = U \ O, any bidder that demands an item from T at
pt−1, is still demanding this item at pt, so D+T (pt−1) ⊆ D+
T (pt). On the other hand, U is
underdemanded at pt, so
|D+U (pt)| < |U |.
Further, observe that H∩D+T (pt−1) = ∅, since H ⊆ D−
O(pt−1) and the members of D−O(pt−1)
demand only items in O, whereas the members of D+T (pt−1) demand at least one item from
T = U \O at pt−1. Therefore, the number of bidders in H that still demand items in S at
pt can be at most |S| − 1. Suppose not, i.e., the number is at least |S|. Then the number
of bidders in D+U (pt) (demanding at least one item of U at pt) is at least equal to |S| plus
the number of bidders in D+T (pt−1), i.e.
|D+U (pt)| ≥ |S|+ |T | = |U ∩O|+ |U \O| = |U |,
contradicting the fact that U is underdemanded. Hence there are at least two bidders in
H that are no longer demanding items in U ∩ O at pt. Select h as one of these bidders
and k as one of the elements in the non-empty set Dh(pt−1) ∩ S. Then item k and bidder
h satisfy the requirements. �
In the special case that U = {k} with k ∈ O, i.e., the single item k in U is overpriced,
we have that no bidder is demanding k at pt. Hence, any bidder h in D−O(pt−1) having k
in his demand set at pt−1 can be selected. Note that according to Corollary 3.3 there are
at least two of such bidders.
In the next lemma we prove that any time when at some stage t + 1 the auction
enters Step 2 after at stage t an item k has been assigned to some bidder h in Step 3,
18
there will be no underdemand of items. So, when the auction arrives in Step 2 after Step
3, either there is neither underdemand nor overdemand and the auction stops, or there
is overdemand and the prices of items in some minimal overdemanded set are increased.
This guarantees that any time when the auction goes to Step 3, prices in some minimal
overdemanded set were increased in the previous stage. Recall that when at stage t + 1
Step 2 is reached from Step 3, the price vector pt+1 is equal to the price vector pt−1, except
that item k has been deleted.
Lemma 4.3 Let U be a minimal underdemanded set at stage t that appears after at stage
t− 1 the prices of the items in a minimal overdemanded set O were increased, and let, in
Step 3, k ∈ U ∩O be the item assigned to some bidder h ∈ {i ∈ M t | Di(pt−1) ⊆ O} such
that k ∈ Dh(pt−1), but k �∈ Dh(pt). When the auction proceeds to stage t + 1 and returns
to Step 2, then there will be no underdemanded set of items.
Proof. First, observe that, by definition of the auction, M t+1 = M t−1 \ {h}, N t+1 =
N t−1 \ {k} and N t+1 �= ∅ (otherwise the auction ends in Step 3). Further, pt+1j = pt−1j for
all j ∈ N t+1. Denote O = O \ {k}. For S ⊆ N t+1 we consider two cases, namely S ⊆ O
and S \ O �= ∅. In the first case we have by Lemma 3.2 that at least |S| + 1 members of
the set D−O(pt−1) = {i ∈ M t−1 | Di(pt−1) ⊆ O} demanded at least one item of S in stage
t− 1. Since pt+1j = pt−1j for all j ∈ N t+1, for any bidder i in M t+1 it holds that
Di(pt+1) = Di(pt−1) \ {k}
and thus any bidder i ∈ M t+1 ∩D−O(pt−1) = D−
O(pt−1) \ {h} that demanded an item of S
at stage t− 1 is still demanding an item of S at stage t + 1. So, when h demanded an item
of S at stage t− 1, the number of bidders of M t+1 demanding an item of S at stage t + 1
is at least |S|, otherwise the number is at least |S|+ 1. Hence S is not underdemanded.
For the second case S \ O �= ∅ we consider the partition of S given by S1 = S ∩ O
and S2 = S \ O. Denote
K1 = {i ∈ D−O(pt−1) | Di(pt−1) ∩ S1 �= ∅}
and
K2 = {i ∈ M t−1 | Di(pt−1) ∩ S2 �= ∅}.
Since Di(pt−1) ⊆ O for all i ∈ D−O(pt−1) and S2 ⊆ N t−1 \ O, it follows that K1 ∩K2 = ∅.
Since O is a minimal overdemanded set at stage t−1 and there is no underdemand at stage
t− 1, we have that S1 is neither overdemanded nor underdemanded at pt−1, because it is
a strict subset of O. By Lemma 3.2 we have that at least |S1| + 1 members of D−O(pt−1)
demanded at least one item of S1 in stage t − 1 and similarly as above it follows that at
19
least |S1| members of D−O(pt−1) \ {h} are still demanding an item of S1 at stage t + 1.
Furthermore, none of these bidders belong to K2, because D−O(pt−1) ∩ K2 = ∅. Further
|K2| ≥ |S2|, because there is no underdemand at stage t− 1. Clearly, any member of K2
is still demanding an item of S2 at stage t + 1, because all prices of the remaining items in
N t+1 are equal to the prices at stage t− 1. Therefore the number of bidders that demand
at least one item of S = S1 ∪ S2 is at least equal to
|S1|+ |K2| ≥ |S1|+ |S2| = |S|
and thus S is not underdemanded at stage t + 1. �
The final lemma states that when in Step 3 an item has been assigned, the new
set of bidders M t+1 cannot become empty. This is obvious when the number of bidders is
bigger than the number of items. However, it also holds when the set of bidders is at most
equal to the number of items. The reason is that when the auction returns from Step 3 to
Step 2 with precisely one bidder, the auction immediately ends in Step 2.
Lemma 4.4 When at some stage t an item k is assigned to some bidder h ∈ M t at Step
3 of the auction, then |M t+1| ≥ 1.
Proof. Each time when an item is assigned in Step 3, the number of items and the
number of bidders decreases with one. Suppose that at some stage t Step 3 occurs for the
kth time. As long as k < |M | − 1 times we have that M t+1 = |M | − k > 1. Now, suppose
that k = |M | − 1 at some stage t. Then M t+1 = 1 and the auction returns to Step 2.
According to Lemma 4.3, there is no underdemand in Step 2. However, because only one
bidder is left, also overdemand cannot occur in Step 2. Hence, according to Lemma 3.7,
a Walrasian equilibrium has been reached for the remaining set of items and bidders and
the auction terminates in Step 2. �
We conclude this section by showing that the auction terminates in a finite number
of stages.
Theorem 4.5 All Steps of the ascending auction are feasible. Moreover the auction
terminates with a feasible allocation in a finite number of stages.
Proof. The auction starts with all prices equal to the seller’s reservation prices. When
there is no overdemand and no underdemand, the auction stops in Step 2. Otherwise,
there is overdemand, because pj = c(j) for all j and thus, by definition, there cannot be
underdemand. Now the prices of all items in a minimal overdemanded set are increased
and Step 2 is repeated until the auction stops or underdemand arises for the first time.
Since the value of any item to any bidder i is finite and any initial endowment mi is also
20
finite, this occurs within a finite number of stages. Then the auction goes to Step 3 and
assigns an item k to some bidder h. By Lemmas 4.1 and 4.2 this step is feasible. After
that the auction stops in Step 3 because all items are assigned or, according to Lemma
4.4 returns to Step 2 with at least one remaining bidder. According to Lemma 4.3 there
is no underdemand when the auction returns to Step 2 after Step 3. Hence, either there is
no overdemand and no underdemand and the auction stops, or there is overdemand again.
Then in a finite number of stages, again one item is assigned in Step 3, or the auction stops
at Step 2. When the auction stops in Step 3, all items are assigned to different bidders
and the auction ends up with an allocation. When the auction stops in Step 2 at some
stage t, according to Lemma 3.7 there is a Walrasian equilibrium allocation with respect to
the remaining items in N t and the remaining bidders in M t and the auction ends up with
an allocation. Since the number of items is finite, the auction terminates with a feasible
allocation in a finite number of stages. �
5 Fundamental properties of the auction
According to Theorem 4.5 the auction finds a feasible allocation in a finite number of
steps. It remains to show that the auction results in a rationed equilibrium. Let π∗ be the
allocation resulting from the auction, i.e., π∗(i) = k for some k ∈ N when bidder i was
assigned an item in either Step 2 or Step 3, and π∗(i) = 0 otherwise; and let p∗ be the
resulting price vector, i.e., when item k is assigned, then p∗k is the price at which item k is
assigned to some bidder h, otherwise p∗k is the price of the item when the auction stops in
Step 2. Since the auction starts with the reservation price vector C, we have that ptk > c(k)
when at stage t item k is assigned in Step 3, ptk ≥ c(k) for all items k ∈ N t when at stage
t the auction stops in Step 2, and pt0 = c(0) = 0 for all t. Hence p∗k = pt−1k = ptk − 1 ≥ c(k)
when item k is assigned at t in Step 3, p∗k = ptk ≥ c(k) for any item k that is assigned at
the final stage t in Step 2 and p∗0 = c(0) and thus p∗ is feasible. Further, when a bidder
gets assigned an item in either Step 2 or Step 3, then the item is in his demand set at
the price the bidder has to pay and thus every bidder i can afford to buy the item π∗(i)
assigned to him. Hence (p∗, π∗) is implementable. We further define the allotment scheme
R∗ as follows. For i ∈M , define Ri∗ by
Ri∗k =
{0 if k ∈ {j ∈ N \ π∗(i) | p∗j ≤ mi and V i(j)− p∗j > V i(π∗(i))− p∗π∗(i)},
1 otherwise.(5.1)
Theorem 5.1 The implementable pair (p∗, π∗) and the allotment scheme R∗ yield a
rationed equilibrium (p∗, π∗, R∗).
21
Proof. We have shown above that (p∗, π∗) is an implementable pair. So, it remains to
prove that the conditions (i)-(v) of Definition 2.2 hold. To prove (i), first consider a bidder
i that got assigned an item k in Step 3 at some stage t against price pt−1k . Then according
to Step 3,
k ∈ Di(pt−1) = {j ∈ N t−1 | pj ≤ mi, V i(j)− pj = max{ℓ∈Nt−1∪{0} | pℓ≤mi}
(V i(ℓ)− pℓ) }
After item k has been assigned to bidder i at stage t, the auction continues with Step 2 at
stage t+1 with the remaining set of items N t+1 = N t−1 \{k}. Since at any stage τ ≥ t+1,
pτj ≥ pt−1j for all j ∈ N t+1, it follows that
V i(k)− p∗k ≥ V i(j)− p∗j , for all j ∈ N t+1 with p∗j ≤ mi.
Further, observe that any j ∈ N \N t−1 has been assigned in some stage τ ≤ t− 1, before
in stage t the item k is assigned to bidder i. According to (5.1) we have that R∗ij = 0 for
all j ∈ N \N t−1 satisfying p∗j ≤ mi and V i(j) − p∗j > V i(k) − p∗k. Hence k ∈ Di(p∗, R∗i).
Second we consider a bidder i who was assigned item k in Step 2 at the final stage t. Such
a bidder i has item k in his demand set Di(pt) with respect to the items in N t. Again, for
any j ∈ N \N t that was assigned before in some stage τ ≤ t−1, we have that R∗ij = 0 when
p∗j ≤ mi and V i(j)−p∗j > V i(k)−p∗k. Hence, also in this case we have that k ∈ Di(p∗, R∗i).
To prove (ii), observe that when an item k is not assigned to a bidder i, then k
belongs to the set N t when the auction stops in Step 2 in the final stage t. According to
Lemma 3.7 then the auction ends with a Walrasian equilibrium allocation with respect to
the remaining items in N t and the remaining bidders in M t. By definition of the Walrasan
equilibrium we then have that p∗k = ptk = c(k) for any unassigned item k.
Since there is a Walrasian equilibrium for the remaining items N t and bidders M t
when at the final stage t the auction stops in Step 2, according to (5.1), rationing only
occurs for items that has been assigned in some Step 3 before the final stage t. This proves
that condition (iii) holds. Further, condition (iv) also follows immediately from (5.1).
To prove (v), again observe that when for some item j we have that π(h) = j and
Ri∗j = 0 for some bidder i �= h, then item j has been allocated at Step 3 before the end
of the auction. Let item j be allocated at some stage t. Then item j was in a minimal
overdemanded set O at pt−1 and for bidder h to which j is assigned it holds that (i)
h ∈ {h′ ∈ M t | Dh′(pt) ⊆ O}, (ii) j ∈ Dh(pt−1) and (iii) for all k ∈ Dh(pt) it holds that
ptk ≤ pt−1j . Since ptk = pt−1k + 1 for all k ∈ O and ptk = pt−1k for all k ∈ N t \ {O}, it follows
that pt−1j = mh, otherwise j should still have been in the demand set of h at pt. Hence
p∗j = pt−1j = mh. This completes the proof. �
As an immediate consequence of the theorem, we have the following corollary.
22
Corollary 5.2 The auction model with financially constrained bidders has at least one
rationed equilibrium.
The next result shows that the rationed equilibrium (p∗, π∗, R∗) generated by the
auction has a particular and interesting property, namely when a bidder is rationed on
some item j, then there exist prices pk at most equal to the equilibrium prices p∗k for all
the other items k �= j, such that at any price pj greater than its equilibrium price p∗j , the
item j does not belong to the unconstrained demand set Dh(p) for every h ∈ M , i.e. at p
no bidder demands item j.
Proposition 5.3 Let (p∗, π∗, R∗) be the rationed equilibrium generated by the auction.
If Ri∗j = 0 for some i, then there exist pk ≤ p∗k for all k �= j, such that j �∈ Dh(p) for any
pj > p∗j and any h ∈M .
Proof. Observe that there is only rationing for items assigned in Step 3. Let item j be
assigned in Step 3 at stage t against price pt−1j . By construction of the auction we have
that pt−1k ≤ p∗k for all k ∈ N t−1. Now, for all k �= j, take pk = pt−1k if k ∈ N t−1 \ {j} and
pk = p∗k if k ∈ N \N t−1. We now consider two cases.
Case A. Item j was overpriced at pt. First consider a bidder h ∈M \M t−1. Such
a bidder has been assigned some item ℓ ∈ N \ N t−1 at some price system pτ , τ < t − 1.
It holds that ℓ ∈ Dh(pτ ). Since p∗j = pt−1j ≥ pτj and p∗ℓ = pτℓ , it follows that j �∈ Dh(p) for
all pj > p∗j = pt−1j , whether or not j was in Dh(pτ ). Second, consider a bidder h in M t−1.
Since j was overpriced at pt, bidder h did not demand j in stage t at price pt. So, since in
stage t only the items in N t were available, we have that j does not belong to the demand
set
Dh(pt) = {ℓ ∈ N t ∪ {0} | ptℓ ≤ mi, V i(ℓ)− ptℓ = max{k∈Nt∪{0} | pt
k≤mi}
(V i(k)− ptk)},
with respect to the set N t of available items at stage t. Since pk = pt−1k ≤ ptk for k ∈
N t−1 \ {j} and pj ≥ ptj = p∗j + 1, it follows that j does not belong to the demand set
Dh(p) = {ℓ ∈ N ∪ {0} | pℓ ≤ mi, V i(ℓ)− pℓ = max{k∈N∪{0} | pk≤mi}
(V i(k)− pk)},
with respect to the set of all items, irrespective of the prices of the items in N \N t.
Case B. Item j was an element of a minimal underdemanded set U t with |U t| ≥ 2.
Similarly as in Case A, for any bidder i ∈ M \ M t−1 we have j �∈ Di(p) for all pj >
p∗j = pt−1j , whether or not j was in Di(pτ ). Second, consider the bidders in M t−1 and let
T = {i ∈M t−1 | Di(pt)∩U t �= ∅} be the set of bidders that demand at pt at least one item
from U t. According to Lemma 3.6, for any bidder in h ∈ T we have that |Dh(pt)∩U t| ≥ 2.
So, for any bidder h ∈ M t−1 it holds that either j �∈ Dh(pt), or j, ℓ ∈ Dh(pt) for some
23
ℓ �= j. In the latter case such a bidder h is indifferent between j and ℓ at pt. This implies
that j will not be chosen by bidder h ∈ T from the set N t when pj > p∗j = pt−1j = ptj − 1
and pℓ = ptℓ − 1. So, bidder h ∈ M t−1 does not demand j at prices pk, k ∈ N t−1 \ {j}
and pj > p∗j when only the items in N t−1 are available. Analogously as in Case A, it
follows that also j �∈ Dh(p), i.e., bidder h ∈ M t−1 also does not have j is his demand set
at p when all items in N available, irrespective of the prices pk of the items k in N\N t−1. �
So far we have considered the case that some or all bidders may confront financial
constraints. We have shown that the proposed ascending auction can handle such a situa-
tion and always finds a rationed equilibrium. One may naturally ask whether the proposed
auction can find a Walrasian equilibrium when no bidder faces a budget constraint. The
following theorem demonstrates that this is indeed the case. More specifically, if every
bidder i is endowed with a sufficient amount mi of money in the sense that mi ≥ V i(j)
for all j ∈ N , then the ascending auction coincides with the DGS auction and finds a
Walrasian equilibrium with the smallest equilibrium price vector in finitely many steps.
Theorem 5.4 If mi ≥ maxj∈N V i(j) for all i ∈ M , then the ascending auction coin-
cides with the DGS auction and finds a Walrasian equilibrium with a minimal equilibrium
price vector p∗ in finitely many steps.
Proof. It is sufficient to show that the ascending auction never generates an underde-
manded set of items in any stage. It is true at stage 1 because the ascending auction starts
with the reservation price vector C. Suppose that at some general stage t, there is no
underdemanded set of items and O is the minimal overdemanded set of items chosen by
the auctioneer as described in Step 2. We will show that there will be no underdemanded
set of items at stage t+1. We first prove that no subset S of the set O is underdemanded at
pt+1. Because mi ≥ maxj∈N V i(j) and 0 �∈ O, every bidder i ∈ D−O(pt) who demands items
from S at pt will continue to demand the same items in S and may demand other items as
well at pt+1. It follows from Lemma 3.2 that the set S cannot be underdemanded at pt+1.
Second, no subset S of N t \O is underdemanded at pt+1, because S is not underdemanded
at pt and the price of each item in N t \O at stage t + 1 is the same as at stage t and the
price of each item in O is increased by one at stage t + 1. Combining the two reasonings
for the case S ⊆ O and S ⊆ NT \O, it follows that also any S ⊆ N t with S ∩O �= ∅ and
S ∩ (N t \O) �= ∅ is underdemanded at pt+1. So the ascending auction will never go to Step
3 and thus coincides exactly with the DGS auction. It is known that their auction finds
an equilibrium with the minimal equilibrium price vector. �
Observe that when the condition of Theorem 5.4 holds, the auction never reaches
Step 3 and thus N t = N in any stage t.
24
6 Conclusion
In this paper we investigated a general and practical market model in which an auctioneer
wants to sell a number of items to a group of financially constrained bidders. Every bidder
knows his values over the items and his budget privately and the auctioneer does not know
this private information unless bidders tell her. When biddders face budget constraints,
a Walrasian equilibrium typically fails to exist. Also the well-known ascending auction
of Demange, Gale and Sotomayor (1986), as well as other auctions, might fail to allocate
all the items. To overcome this the concept of rationed equilibrium has been proposed.
Moreover and most importantly, an ascending (open) auction is developed which, starting
with the seller’s reservation price of each item, always ends up with a rationed equilibrium
in finitely many steps. This makes the auction an attractive allocation mechanism in
situations with financially constrained bidders. By using the auction the seller may increase
his revenues and the total value in the market may increase compared to the disequilibrium
situations in which not all items are allocated. Furthermore, because the auction yields a
rationed equilibrium, it implies that a rationed equilibrium always exists in the assignment
market with financial constraints. This result is parallel to the celebrated result obtained
by Shapley and Shubik (1972), who proved the existence of a Walrasian equilibrium in
the assignment market without financial constraints. In contrast to their method, our
approach is constructive and gives us an exact solution. We have also shown that when no
bidder is financially constrained, the proposed auction reduces to the auction of Demange,
Gale and Sotomayor (1986).
References
[1] C. Azariadis and J. Stiglitz (1983), “Implicit contracts and fixed price equilibria,”
Quarterly Journal of Economics, 98, 1-22.
[2] J.P.Benoıt and V. Krishna (2001), “Multiple-object auctions with budget constrained
bidders,” Review of Economic Studies, 68, 155-179.
[3] Y.K.Che and I. Gale (1996), “Expected revenue of the all-pay auctions and first-price
sealed-bid auctions with budget constraints,” Economics Letters, 50, 373-380.
[4] Y.K.Che and I. Gale (1998), “Standard auctions with financially constrained bidders,”
Review of Economic Studies, 65, 1-21.
[5] C. Cox (1980), “The enforcement of public price controls,” Journal of Political Econ-
omy, 88, 887-916.
25
[6] V.P. Crawford and E.M. Knoer (1981), “Job matching with heterogeneous firms and
workers,” Econometrica, 49, 437-450.
[7] G. Demange, D. Gale, and M. Sotomayor (1986), “Multi-item auctions,” Journal of
Political Economy, 94, 863-872.
[8] J.H. Dreze (1975), “Existence of an exchange economy under price rigidities,” Inter-
national Economic Review , 16, 310-320.
[9] F. Gul and E. Stacchetti (1999), “The English auction with differentiated commodi-
ties,” Journal of Economic Theory, 92, 66-95.
[10] P. Hall (1935), “On representitives of subsets,” Journal of London Mathematical So-
ciety , 10, 26-30.
[11] G. Illing and U. Kluh (2003), Spectrum Auctions and Competition in Telecommunica-
tions, MIT Press, London.
[12] P. Klemperer (2004), Auctions: Theory and Practice, Princeton University Press,
Princeton, New Jersey.
[13] T.C. Koopmans and M. Beckmann (1957), “Assignment problems and the location of
economic activities,” Econometrica, 25, 53-76.
[14] V. Krishna (2002), Auction Theory, Academic Press, New York.
[15] M. Kurz (1982), “Unemployment equilibria in an economy with linked prices,” Journal
of Economic Theory, 26, 110-123.
[16] G. van der Laan (1980), “Equilibrium under rigid prices with compensation for the
consumers,” International Economic Review, 21, 53-73.
[17] J.J. Laffont and J. Roberts (1996), “Optimal auction with financially constrained
buyers,” Economics Letters, 52, 181-186.
[18] E. Maskin (2000), “Auctions, development, and privatization: efficient auctions with
liquidity-constrained buyers.” European Economic Review, 44, 667-681.
[19] D. Mishra and D. Talman (2006), “Overdemand and underdemand in economies with
indivisible goods and unit demands,” DP No.2006-84, CentER for Economic Research,
Tilburg University, Tilburg.
[20] D. Mishra and D. Veeramani (2006), “An ascending price procurement auction for
multiple items with unit supply,” IEE Transactions, 38, 127-140.
26
[21] T.R. Palfrey (1980), “Multiple object, discriminatory with bidding constraints: a
game-theoretic analysis,” Management Science, 26, 935-946.
[22] C. Pitchik and A. Schotter (1988), “Perfect equilibria in budget-constrained sequential
auctions: an experimental study,” Rand Journal of Economics, 19, 363-388.
[23] M. Rothkopf (1977), “Bidding in simultaneous auctions with a constraint on expo-
sure,” Operations Research, 25, 620-629.
[24] D. Salant (1997), “Up in the air: GTE’s experience in the MTA auction for personal
communications services licenses,” Journal of Economics and Management Strategy,
6 (3), 549-572.
[25] L.S. Shapley and M. Shubik (1972), “The assignment game I: the core,” International
Journal of Game Theory, 1, 111-130.
[26] M. Sotomayor (2002), “A simultaneous descending bid auction for multiple items and
unitary demand,” Revista Brasilerira de Economia Rio de Janeiro, 56, 497-510.
[27] D. Talman and Z. Yang (2006), “A dynamic auction for differentiated items under
price rigidities,” DP No.2006-26, CentER for Economic Research, Tilburg University,
Tilburg, forthcoming in Economics Letters.
[28] W. Vickrey (1961), “Counterspeculation, auctions, and competitive sealed tenders,”
Journal of Finance, 16, 8-37.
27